Published in ACM Transactions on Graphics (SIGGRAPH 2003) Layered Acting For Character Animation Mira Dontcheva Gary Yngve Zoran Popovic´ University of Washington Figure 1: While watching a real-time display of her actions, an animator performs a hopping motion, sketching the trajectory of a kangaroo. Abstract allows expert actors to animate human characters, but is fraught with a different set of difficulties, particularly when animating non- We introduce an acting-based animation system for creating and human characters and when attempting to edit existing motion. We editing character animation at interactive speeds. Our system re- present an interface that supports the many benefits of performance quires minimal training, typically under an hour, and is well suited animation yet allows for the mapping betweeen the animator and for rapidly prototyping and creating expressive motion. A real-time character to be established in ways that are both flexible and easy motion-capture framework records the user's motions for simulta- to understand. neous analysis and playback on a large screen. The animator's real- We had several goals in mind when designing our system. We world, expressive motions are mapped into the character's virtual wanted our system to have an easy-to-use and efficient interface ap- world. Visual feedback maintains a tight coupling between the an- propriate for a novice with little training. We wanted to give the user imator and character. Complex motion is created by layering mul- control and the ability to create motions that have a sense of person- tiple passes of acting. We also introduce a novel motion-editing ality, often lacking in 3-d animation. Our results and user studies technique, which derives implicit relationships between the anima- indicate that we have made progress towards achieving these goals. tor and character. The animator mimics some aspect of the charac- ter motion, and the system infers the association between features We introduce a system that allows users to create and edit char- of the animator's motion and those of the character. The anima- acter animation by acting. A motion capture system and a set of re- tor modifies the mimic by acting again, and the system maps the flective props, or widgets, provide the connection between the actor changes onto the character. We demonstrate our system with sev- and the character, as shown in Figure 1). The motion capture system eral examples and present the results from informal user studies frees the animator from the confines of a mouse, a keyboard, and with expert and novice animators. a limited workspace. The user animates by acting while watching a large display for instant feedback. The system requires minimal Keywords: Character Animation, Motion Editing, Statistical training, typically under an hour, because it innately captures users' Analysis, 3D User Interfaces, Motion Transformation expressiveness from their acting. The key contributions of our work lie in how the animator's mo- 1 Introduction tions are mapped to those of the character, and how animation can be built upon layer by layer. The mapping from actor to character is Animation should come from the heart, not from the head. Current created by a combination of explicit and implicit controls. Explicit modeling and animation packages favor control over ease of use control is used for rapid prototyping of the initial animation. In each and demand a level of expertise from the user that is out of reach layer, multiple related degrees of freedom (DOFs) of the character for all but a few highly skilled animators. Direct motion capture are modified simultaneously. Layering allows the animator to fo- cus on one aspect of the animation at a time, whereas working with all aspects at once can be overwhelming. When editing, implicit control can be created by automatically inferring which DOFs the animator is trying to control. The system implicitly creates the map- ping of real-world motion to that of the character for editing; the mapping may involve both spatial and temporal transformations. The rest of the paper proceeds as follows. First, we discuss rele- vant related work and give an overview of the system and the inter- face. We then provide details of the algorithms for motion creation, motion editing with explicit transformations, and inferred motion editing with implicit transformations. Finally, we conclude with results, user studies, and future work. 1 Published in ACM Transactions on Graphics (SIGGRAPH 2003) 2 Related Work The research presented in this paper was inspired by previous contributions to interactive and easy-to-learn interfaces, especially those intended for novices. Existing work includes sketching out- lines into 3-d shapes [Igarashi et al. 1999], sculpting surfaces, [Schkolne et al. 2001], and using plush toys for interaction with virtual environments [Johnson et al. 1999]. Many animators rely on explicit keyframing, where at a specific instant in time, they have full control of a character. The difficulty comes when animators want to express dynamic, fluid motion, par- ticularly when multiple degrees of freedom should be manipulated in unison. We allow the user to edit multiple DOFs over a period of time, and because time is inherent in the process of acting, rather than in an axis of a graph, we retain the intuition that is lacking from keyframes and graph editors. Motion capture has a long history in computer animation as a Figure 2: An animator is interacting with our system. The real- mechanism for transferring live motion to the digital domain. Mo- time motion-capture system tracks the widget in her hand (the user tion capture can record the nuances that give feeling and expres- feature). The widget's motion, X(t), maps to a character feature, siveness to motion; however, it also has several limitations. The Y(t), and the user's edits are instantly reflected on the screen. computer-generated character may not have the same physique or dimensions as the human actor. The desired motion may be too dangerous to act and capture in real life or may involve impossi- ble motions requiring superheroic speed or strength. We rely on mation system, which combined direct manipulation of DOFs with motion-capture technology to support our work, but we do not tie physical controls as guides for more natural joint motion. We ex- the character's motion directly to that of the actor. This indirect tended their layering to include additive layering, which allows for interaction links us closer to the realm of puppeteering. refinement of existing motion. In addition, our system allows not Puppeteering and performance animation have been shown to be only direct manipulation of DOFs but also indirect manipulation of powerful tools for the expert user. An animator uses a specialized relative trajectories through the use of inverse kinematics. We feel input device, ranging from an exoskeleton to a force-feedback sty- that the layering and indirect interaction make our system easier for lus, to control a character through some direct mapping of DOFs. users, especially novices. Personality is imparted directly onto the character from the nature There are other related interactive interfaces for animation. of the control. Our work follows on this work with the goal of en- Popovic,´ et al. create precise animations of rigid bodies [Popovic´ abling novice users to easily create expressive animation. et al. 2000]. Perlin has a simulated actor that responds based on human intervention [Perlin 1995]. Laszlo et al. directly involve Although the earliest computer puppetry work dates back to the the animator in the control of interactive physics-based simula- late sixties [Sturman 1998], modern performance animation traces tions of walking, jumping, and climbing using mixes of continuous back to the Jim Henson Company [Walters 1989] and deGraf and (mouse-based) and discrete (keypress-based) control [Laszlo et al. Wharman [de Graf 1989]. Following the debut of Mike the Talk- 2000]. Donald and Henle use haptics to construct new motion by ing Head at Siggraph 89, Protozoa [de Graf and Yilmaz 1999], editing low-dimensional representations of existing motion [Donald Medialab [Benquey and Juppe 1997], and others deployed per- and Henle 2000]. Several systems use gestures to control a perfor- formance animation systems on television programs, showing the mance, including the Personal Orchestra [Borchers et al. 2001] and power of using acting for animation. Our work differs in several Andy Wilson's Seagull from the 1996 Siggraph Digital Bayou. ways. Whereas existing systems use direct or preassigned map- pings, we generalize the mapping between the actor and charac- ter, granting flexibility with the character and the motion-capture rig. We animate parts of the character individually using separate 3 Overview of System motion-capture takes, whereas traditionally, most parts of a charac- ter are animated simultaneously. Unlike existing performance ani- In this section, we give an overview of the system from the hard- mation systems, ours requires minimal calibration, making our tool ware and software perspectives. useful for rapid prototyping and accessible to novices. Several researchers have investigated computer puppetry. Shin, Hardware Our system uses an 8-camera optical motion-capture et al. developed a computer puppetry system allowing a performer system, a wireless mouse, a wall display screen, and any number of to animate an entire character online [Shin et al. 2001]. They infer animation widgets (see Figure 2). The system interprets each wid- context from the motion, which they use to guide inverse kinemat- get abstractly as a user feature. We gather motion-capture data at ics. Our goals differ, as they want a fully markered puppeteer to 120 fps and compute the translational and rotational values of each create instant motion for a retargeted character, but we want an ani- widget in real time.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages8 Page
-
File Size-